Perez and colleagues: language-model-generated red-team tests

What changed

Used language models to generate adversarial test inputs and compared generation and learning strategies for finding undesirable target-model outputs, offering automated discovery as a complement to manual red-teaming.

Scientific scope

Used language models to generate adversarial test inputs and compared generation and learning strategies for finding undesirable target-model outputs, offering automated discovery as a complement to manual red-teaming. The search inherits biases from its generator and harm classifier, with trade-offs between diversity and judged severity. Existing supervised and reinforcement-learning methods are components, not all new inventions of this work.

Evidence

The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. The search inherits biases from its generator and harm classifier, with trade-offs between diversity and judged severity. Existing supervised and reinforcement-learning methods are components, not all new inventions of this work.

Contributors

• The named authors in the research record · SCIENCE: Source-listed authors for the exact contribution; experimental subjects are not autonomous scientific authors. Source support and predecessor work retain separate roles. Individual effort shares and unresolved identities are not inferred.

• Red Teaming Language Models with Language Models

Provisional WRS assessment, edition 2026-10-23.4. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction